An adaptive segmentation scheme based on recurring action potentials for sEMG controlled movement decoding
Anil Sharma1, Nikhil Vivek Shrivas2, Ila Sharma3
1Department of Electronics and Communication Engineering, Malaviya National Institute of Technology, Jaipur, India. 2020rec9510@mnit.ac.in.
Physical and Engineering Sciences in Medicine
|May 6, 2025
Summary
This study introduces an adaptive segmentation method for electromyography (EMG) signal processing, improving feature extraction for better decoding accuracy. The novel approach enhances system performance compared to traditional constant width segmentation.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Machine Learning
Background:
- Electromyography (EMG) controlled decoding systems are crucial for prosthetics and human-computer interfaces.
- Conventional EMG signal processing relies on constant width segmentation, which struggles with the inherent complexity and randomness of EMG signals.
- There is a need for more sophisticated segmentation techniques to improve the accuracy and reduce the delay in EMG-based systems.
Purpose of the Study:
- To propose and validate a novel adaptive segmentation approach for EMG signal processing.
- To enhance feature extraction by capturing action potential patterns for improved decoding accuracy.
- To compare the performance of the proposed adaptive segmentation with conventional constant width segmentation.
Main Methods:
- Developed a novel adaptive segmentation method based on the repeating patterns of action potentials in EMG signals.
- Extracted twenty time-domain features from segmented EMG data.
- Employed Linear Discriminant Analysis (LDA), k-nearest neighbor (kNN), and Decision Tree (DT) classifiers to evaluate performance.
- Experimentally validated the approach with 12 subjects performing eight distinct movements.
Main Results:
- The proposed adaptive segmentation achieved an average segmentation width of 124 ms with a narrow margin of error.
- Average F1 scores across subjects and movements were 82.078% (LDA), 81.51% (kNN), and 80.81% (DT).
- 5-fold cross-validated accuracies reached 78.3% (LDA), 78.2% (kNN), and 76.70% (DT).
- Statistical analysis (t-test) indicated significant performance improvements over constant width segmentation.
Conclusions:
- The proposed adaptive segmentation method effectively captures EMG signal complexities, outperforming constant width segmentation.
- This novel approach offers a promising solution for enhancing the accuracy and efficiency of EMG-controlled decoding systems.
- The adaptive segmentation strategy provides a more robust foundation for feature extraction in real-time EMG applications.
